A distributed-commonsensical approach for human activity recognition is proposed.
The approach relies on video and sensor analysis enriched with common sense.
Sequences of spatio-temporal actions are interpreted as causal explanations.
Parallel causal explanations are tracked although just one is eventually considered.